Vocalization patterns of dairy animals to detect animal state

Animals cannot communicate the different states of their being - such as normal, hunger, or heat state - through semantics. However, they do generate voices in different states. In this paper, we start with the hypothesis that identification of the specific state of the animal is possible by analyzing their speech signals. We use a variety of spectral features for the purpose of identifying the type of a dairy animal, and then the state of a particular animal. The animal vocalization data is collected through regular microphones and the audio is then analyzed by extracting features. The details of the data collection process, feature extraction and classification results are presented in this paper. Experiments performed on 60 animals provide a strong argument for the usefulness of the vocalization pattern analysis techniques for animal identification and state detection. The paper therefore paves a new direction for non-intrusively detecting the state in dairy animals.